Veröffentlichungsversion / Published Version Arbeitspapier / working paper

Transcription

1 More than wishful thinking : causes and consequences of voters electoral expectations about parties and coalitions Meffert, Michael F.; Huber, Sascha; Gschwend, Thomas; Pappi, Franz Urban Veröffentlichungsversion / Published Version Arbeitspapier / working paper Zur Verfügung gestellt in Kooperation mit / provided in cooperation with: SSG Sozialwissenschaften, USB Köln Empfohlene Zitierung / Suggested Citation: Meffert, Michael F. ; Huber, Sascha ; Gschwend, Thomas ; Pappi, Franz Urban : More than wishful thinking : causes and consequences of voters electoral expectations about parties and coalitions. Mannheim, URN: Nutzungsbedingungen: Dieser Text wird unter einer Deposit-Lizenz (Keine Weiterverbreitung - keine Bearbeitung) zur Verfügung gestellt. Gewährt wird ein nicht exklusives, nicht übertragbares, persönliches und beschränktes Recht auf Nutzung dieses Dokuments. Dieses Dokument ist ausschließlich für den persönlichen, nicht-kommerziellen Gebrauch bestimmt. Auf sämtlichen Kopien dieses Dokuments müssen alle Urheberrechtshinweise und sonstigen Hinweise auf gesetzlichen Schutz beibehalten werden. Sie dürfen dieses Dokument nicht in irgendeiner Weise abändern, noch dürfen Sie dieses Dokument für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, aufführen, vertreiben oder anderweitig nutzen. Mit der Verwendung dieses Dokuments erkennen Sie die Nutzungsbedingungen an. Terms of use: This document is made available under Deposit Licence (No Redistribution - no modifications). We grant a non-exclusive, nontransferable, individual and limited right to using this document. This document is solely intended for your personal, noncommercial use. All of the copies of this documents must retain all copyright information and other information regarding legal protection. You are not allowed to alter this document in any way, to copy it for public or commercial purposes, to exhibit the document in public, to perform, distribute or otherwise use the document in public. By using this particular document, you accept the above-stated conditions of use.

2 : Causes and Consequences of Voters Electoral Expectations about Parties and Coalitions Michael F. Meffert Department of Political Science Leiden University Wassenaarseweg AK Leiden Netherlands Phone: Fax: Thomas Gschwend Graduate School of Economic & Social Sciences (GESS) Universität Mannheim Mannheim Germany Phone: Fax: Sascha Huber Mannheimer Zentrum für Europäische Sozialforschung (MZES) Universität Mannheim Mannheim Germany Phone: Fax: Franz Urban Pappi Lehrstuhl für politische Wissenschaft I Universität Mannheim Mannheim Germany Phone: Fax: fupappi@rumms.uni-mannheim.de

3 1 : Causes and Consequences of Voters Expectations about Election Outcomes Abstract Accurate expectations about the outcome of elections play a central role in psychological and economic theories of voting. In the paper, three questions about voters expectations are investigated. First, we identify and test several factors that influence the overall accuracy or quality of voters expectations. Second, the phenomenon of wishful thinking is tested and confirmed for expectations about the electoral performance of individual parties and coalitions. Finally, two mechanisms how expectations might influence voting behavior are identified and tested. Based on surveys from Austria and Germany, the results suggest that voters not only rely on expectations to avoid casting wasted votes for parties without electoral chances, but that they are able to engage in fairly sophisticated strategic coalition voting.

4 2 The media coverage of polls during political campaigns is extensive, giving even voters without much interest in politics an opportunity to learn rather sophisticated information about an upcoming election (Brettschneider, 2000; 2003). If voters choose to do so, they can use this readily available information to form fairly accurate expectations. In economic theories of voting (Cox, 1997; Downs, 1957), these expectations are a critical and unquestioned assumption. Similar assumptions are made, explicitly or implicitly, by psychological theories such as the spiral of silence (Noelle-Neumann, 1993) or the bandwagon and underdog effect (Mutz, 1998; Simon, 1954). Empirical research of voters electoral expectations, however, takes a rather skeptical view of the claim that voters are able form highly accurate expectations. Voters with strong (partisan) preferences tend to engage in wishful thinking and overestimate the chances of preferred parties and candidates and/or underestimate the chances of disliked parties and candidates (Mutz, 1998). As a consequence, voters expectations appear to be a mix of objective, factual poll information and preference-driven projections (e.g. Blais and Bodet, 2006). The paper addresses three questions. First, we identify and test several factors that influence the overall accuracy and quality of voters expectations. Second, the phenomenon of wishful thinking is tested and confirmed for expectations about the electoral performance of individual parties and extended to judgments about coalitions. Finally, two mechanisms how expectations might influence voting behavior are identified and tested, in particular strategic voting and the bandwagon effect. The analyses are based on two general population surveys from Austria and Germany that measured voters expectations in various ways. The following review will focus first on the sources of voters expectations, followed by the consequences for voting behavior. 1. Sources of Voters Electoral Expectations

5 3 The formation of meaningful expectations about electoral outcomes requires current and precise information. What might appear to be a challenging task, given the well-known low levels of factual political knowledge of many voters (Zaller, 1992), can actually be accomplished rather easily. The media coverage of political campaigns, in particular at the national level, spends considerable time reporting results and trends based on frequent pre-election polls (Brettschneider, 2000; 2003). The reality, of course, is more complex. First, even professional polls do not always accurately predict the election outcome (as was the case in the Austrian and German general elections analyzed here). Second, the (German) media coverage is dominated by subjective claims and assessments by journalists and politicians that are not constrained in any way by professional polls (Donsbach and Weisbach, 2005). As a starting point, it is nevertheless reasonable to assume that voters have fairly easy access to objective and for the most part fairly accurate information about the electoral chances of parties during political campaigns. The more interesting question is about the reception part whether and how voters actually acquire this information to form accurate expectations. According to the pertinent literature, starting with the classic study The People s Choice (Lazarsfeld et al., 1948), voters expectations frequently seem to follow a different logic. Despite the ready access to objective poll information, the literature finds fairly consistent wishful thinking effects, that is, perceptions distorted by existing political preferences. Voters (like sports fans) seem to engage in strong and consistent wishful thinking (Abramson et al., 1992; Babad et al., 1992; Babad and Yacobus, 1993; Granberg and Brent, 1983; Uhlaner and Grofman, 1986). In addition, wishful thinking appears to be one of the few effects that are reliably found in survey research studies but that are very difficult to recreate in laboratory settings. Carefully designed experiments often fail to show any remarkable wishful thinking effects (e.g. Bar-Hillel and Budescu, 1995). Price (2000) suggests a number of explanations, in particular that people s social

6 4 interactions in real life are highly selective, and that the (induced) preferences in the laboratory are not sufficient to produce wishful thinking effects. In his own analysis, Price (2000) shows that the latter problem can be addressed by a carefully designed desirability manipulation (involving two competing social groups). Because the analyses reported in our paper draws on partisan preferences in real world settings, the presence of wishful thinking can be considered as fairly certain and similar to previous studies. But how can the fact that voters engage in wishful thinking despite easy access to accurate poll information be reconciled? It will be useful to differentiate between the overall accuracy or quality of voters expectations and the tendency to distort the expectations for specific preferred and disliked parties. The overall quality of expectations should benefit from ready access to poll information, while not precluding distortions for specific parties. Once strong partisan predispositions come into play (and polls that contradict preferred outcomes), factual information will quickly lose its luster. The literature offers many explanations that can be narrowed down to a number of factors that might explain the overall quality of voters expectations and/or the tendency to engage in wishful thinking. The following review categorizes the factors in political motivations such as partisan preferences and non-partisan political knowledge, rational or strategic considerations, and social context. 1.1 Partisan and Nonpartisan Political Motivations Partisan preferences, in particular party identification, exert a powerful influence over political attitudes and perceptions (Bartels, 2002). Thus, it is hardly surprising that voters expectations about electoral outcomes should be affected by these preferences. A partisan preference implies a strong directional motivation that favors preferred outcomes or parties over disliked outcomes or parties. Psychologically, both motivational and cognitive mechanisms have

7 5 been proposed to explain this self-serving misperception (Babad, 1995; 1997; Bar-Hillel and Budescu, 1995; Price, 2000). Granberg and Brent (1983) favor Heider s (1958) balance theory as explanation for wishful thinking. Because the surveys used in the analyses below do not allow a test of the precise psychological mechanisms, the review will not address this issue in more detail. The more relevant question is whether partisan preferences might have non-partisan implications as well. Voters with a strong party identification and clearly defined political preferences more general exhibit a higher degree of political interest and involvement than voters without these convictions. All else being equal, a partisan voter is invested in the political system, and as a member of the polity (Lewis-Beck and Skalaban, 1989: 153) likely to be familiar with the parties and their approximate electoral strengths. If these claims are correct, partisan preferences should have two distinct effects. First, they should play a central role in wishful thinking, distorting the expectations for preferred and disliked parties. But second, they should also have a positive effect on the overall accuracy or quality of voters expectations due to stronger political involvement. Compared to partisan preferences, the effect of non-partisan political motivations such as political interest and political knowledge is much easier to describe. Without any directional partisan implications, interest in politics and the campaign as well as the cognitive ability to process political information more efficiently should have positive effects on the quality of voters expectations. Political knowledge in particular has been found to play a crucial role in the acquisition of political information in general (Price and Zaller, 1993; Zaller, 1992). It also improves the accuracy of forecasts and lowers the tendency of biased information processing such as wishful thinking (e.g. Lemert, 1986; Babad, 1997; Dolan and Holbrook, 2001; Meffert and Gschwend, 2007; Uhlaner and Grofman, 1986; Yaniv et al., 2002; but see Babad, 1995).

8 6 Voters with a high level political knowledge should be more receptive to political information such as polls and better able to store and retrieve such information from memory. 1.2 Rational and Strategic Considerations From the perspective of a rational voter, the main goal of a vote decision is to maximize expected utility basically the benefit derived from the policy output of the next government (Downs, 1957; Riker and Ordeshook, 1968). In order to maximize expected utility, a rational voter cannot rely on policy and partisan preferences alone but has to take into account the expected outcome of the next election. Given these expectations, a voter might expect a higher benefit by deserting the most preferred party if it has a low chance of winning. By casting a strategic vote for a less-preferred choice with better chances, the voter is more likely to obtain a desirable outcome (Cox, 1997; Fisher, 2004). Thus, strategic voters need poll information to form accurate expectations. The cost of acquiring readily available poll information is, after all, very low. The extent to which a rational voter needs such information to form highly accurate expectations should depend on the difficulty and uncertainty of the decision. If, for example, a voter prefers two or more parties to a similar degree (that is, would obtain the same utility from both parties), the decision will be more difficult and depend on the electoral chances of the parties, increasing the need for precise poll information. Under these circumstances, the expectations about electoral chances will determine which party will maximize the expected utility. Larcinese (2007), for example, reports evidence that voters in constituencies with close contests have a higher demand for campaign information. Finally, and rather obviously, only voters who intend to cast a vote should have a need to form highly accurate expectations. The usefulness of this information will be much lower for those who plan to abstain. 1.3 Social Context

9 7 The literature on wishful thinking offers another, very different explanation why the expectations of individual voters might differ from the national average. If voters live in regions, constituencies, or states that differ politically from the national level (Uhlaner and Grofman, 1986), or if they are embedded in politically homogenous personal networks (Fischer and Budescu, 1995), they might encounter only a biased sample of political opinions that they mistakenly extrapolate to the national level. Fischer and Budescu (1995), for example, suggest that Israeli voters have only selective social interactions that result in distorted inferences about electoral support for their candidates and parties. These approaches assume face-to-face contact. However, it might be that case that mediated expressions have similar effects. Daschmann (2000) shows that voter statements in the media have a larger effect on the perceived climate of opinion than poll results. Because polls have become such a prevalent feature of media coverage in national campaigns, it would be reasonable to expect that social context plays a lesser role than it used to do, especially in the study by Uhlaner and Grofman (1986). But if considerable regional differences exist, for example when states in a federal system are dominated by different parties, it is nevertheless still plausible that voters find themselves in biased social contexts. In fact, accurate perceptions in or of biased contexts might lead to biased out-of-sample forecasts. 2. Electoral Consequences of Voters Expectations The formation of expectations about electoral outcomes is an interesting topic in itself, but the real significance of voters expectations, whether accurate or distorted, derives from their electoral consequences. As documented by Mutz (1998), there are numerous areas of research that have investigated this question, including momentum in American presidential primaries (Bartels 1988), the effect of publicized exit polls (and pre-election polls more generally) on election day (Sudman, 1986), and assorted other theories such as the spiral of silence (Noelle-

10 8 Neumann 1993) and the third-person effect (Pan et al., 2006). These approaches share the assumption that the perception of others opinions will have direct consequences on attitudes and behaviors. Despite the extensive interest and amount of research, the evidence for most of these theories is at best mixed (Mutz, 1998). Mutz has nevertheless identified a number of promising pathways how perceptions of mass opinion might influence (impersonally) individual attitudes and behavior. Two of these, strategic voting and the bandwagon effect, have considerable support and are of particular interest here. 2.1 Strategic Voting If voters are rational actors that maximize their expected utility, they have to take into account the electoral chances of parties and candidates. The most obvious strategy is to avoid a wasted vote for a party or candidate without any electoral chances. Beyond this basic motivation, the vote becomes only strategic if the beneficiary is a party other than the most preferred party, and one that will produce a more desirable outcome, for example a coalition government (Blais et al., 2006). Needless to say, such decision processes require considerable information and political sophistication. Only voters with a high level of political knowledge should be able to engage in such strategic (coalition) voting. Given these high requirements, it is necessary to address and justify the relevance of this decision strategy. Strategic voting has been documented mostly for political systems with majoritarian elections, in particular the British system with its single member districts. However, there is evidence that strategic voting also happens in parliamentary systems using proportional representation. In both cases, the number of strategic voters in representative surveys is relatively low, ranging between 5 and 15 percent (Fisher, 2004). This share is misleading for a number of reasons. First, it refers only to actual strategic voters (or realized strategic votes), excluding all those who might have considered a strategic vote but decided against it. After all, the cognitive

11 9 decision process of strategic voting cannot be observed directly. Second, a fairly small number of voters is sufficient to have a decisive impact in close elections, giving strategic voters a disproportionate influence. Third and most important, strategic voting will only happen if the appropriate incentives and opportunities are given. For example, voters must have plausible alternative choices. By focusing only on those with an opportunity to vote strategically, Alvarez, Boehmke, and Nagler (2006) have shown that the share of strategic voters increases dramatically. A similar analytical strategy will be used below. 2.2 The Bandwagon Effect A second promising pathway of influence identified by Mutz (1998) is the classic bandwagon effect. It is very simple and merely assumes that people want to follow the winner. This mechanism has been used to explain momentum in presidential primaries, the phenomenon that early winners and/or a front-runner status in the polls creates a dynamic that draws uncommitted voters to the winning candidate (Bartels, 1988). While the evidence from presidential primaries is by far the strongest, there is also evidence of more limited bandwagon effects in a various other countries (Gimpel and Harvey, 1997; Lanoue and Bowler, 1998; McAllister and Studlar, 1991; Nadeau et al., 1994). As Mutz (1998) points out, the mechanism of the bandwagon effect considers perceptions of mass opinion as a simple heuristic cue. Basing a vote decision on such a social cue will be considered irrational by proponents of the rational voter paradigm. It is nevertheless clear that voters without much political knowledge or strong partisan predispositions should be more susceptible to this kind influence while voters with a high level knowledge should be more immune. 3. Data and Methods 3.1 Data

12 10 Two nationally representative pre-election surveys from Austria and Germany were used to test the hypotheses. The Austrian survey was conducted September 19 30, 2006, ending the day before the general election for the Austrian Nationalrat on October 1, A nationally representative sample of 1501 respondents was interviewed by phone. The German survey was conducted August 8 to September 17, 2005, ending the day before the general election for the German Bundestag on September 18, A representative sample of 3583 respondents was interviewed by phone. Both surveys include a number of similar measures for party and coalition preferences as well as electoral expectations. The subsequent analyses will report similar models for both countries that differ only for very few variables that are unique to each data set. Before discussing the measures in more detail, some background information for each election will be helpful. 3.2 The Election Context in Austria and Germany At the beginning of the 2006 election campaign for the Austrian Nationalrat, six parties had reasonable chances of obtaining seats in the next parliament (Müller, 2008; Pappi, 2007). These included the two large parties, the governing conservative People s Party (ÖVP) and its challenger, the Social Democrats (SPÖ). Among the smaller parties, the nationalist and populist Freedom Party (FPÖ) and the environmental Greens (Die Grünen) were not only two well established parties but also expected to do very well, likely reaching 10 or more percent. Two other small parties were fairly new. The Alliance for the Future of Austria (BZÖ) was founded in the spring of 2005 by former members of the FPÖ which included all FPÖ ministers in the coalition government with the ÖVP and most FPÖ members in parliament. The BZÖ effectively replaced the FPÖ as the junior coalition partner of the ÖVP. This, however, did not lead to an electoral advantage. The polls gave the BZÖ only minor chances to pass the Austrian minimum vote threshold of 4 percent. The other new party, Liste Dr. Martin, was founded by an

13 11 independent member of the European Parliament, mostly as a protest against the established parties. The polls gave him a reasonable chance of passing the minimum vote threshold. Based on the polls, the unpopular incumbent coalition of ÖVP and BZÖ was expected to lose its majority, but the ÖVP was still expected to stay ahead of the SPÖ by a few percentage points. With BZÖ and Martin close to the 4% threshold, the outcome of the election was fairly uncertain. The polls, however, missed the election outcome. The SPÖ (35.3%) ended up with the largest vote share, beating the unexpectedly weak ÖVP (34.3%) by a small margin. As expected, Greens (11.0%) and FPÖ (11.0%) performed very well. The BZÖ (4.1%) performed better than expected and (barely) managed to pass the minimum vote threshold. Martin (2.8%) clearly failed to gain the necessary support. The results gave only one two-party coalition an absolute majority, a grand coalition of SPÖ and ÖVP, which eventually agreed to form the new government. In Germany, the governing coalition of Social Democrats (SPD) and Green Party (Bündnis 90/Die Grünen) called the next general election for the Bundestag one year early after suffering a significant loss in a crucial state election in May According to the polls, the two main opposition parties, Christian Democrats (CDU/CSU) and the liberal Free Democrats (FDP), had a reasonable chance to win the upcoming election. In addition, the newly constituted Left Party (Linkspartei) introduced considerable uncertainty in the campaign. The Left Party was the result of a merger of the PDS (the successor party of the former communist party in East Germany) and the WASG (a fairly new party in the Western part of Germany drawing disaffected and/or former members of labor unions and SPD). In the polls, this party surpassed both FDP and Greens during the summer. Because no other party was willing to form a coalition with this party, a strong showing was likely to prevent either SPD and Greens or CDU and FDP from forming a coalition government, forcing the formation of either a rather unlikely and unwieldy three-party coalition or a grand coalition between CDU and SPD.

14 12 Similar to Austria, the polls performed poorly in predicting the election outcome. The CDU (35.2%) lost about five percentage points compared to the most recent polls before the election, barely staying ahead of the SPD (34.2%). The FDP (9.8%) performed much better than predicted, followed by Left Party (8.7%) and Greens (8.1%). Given the strong showing of the Left Party, CDU and SPD eventually formed a grand coalition. 3.3 Measurement of Preferences and Expectations The two key measures to investigate the issue of wishful thinking are partisan preferences and respondents expectations about the electoral outcome. Starting with the preferences, it is important to keep in mind that both counties have multiparty systems that defy a simple classification of respondents on a unidimensional party identification scale. Respondents in both surveys rated the relevant parties in each country on 11-point evaluation scales, ranging from -5 ( don t like the party at all ) to +5 ( like the party very much ). While these party evaluations can be used directly as a measure of party preference, an additional dichotomous party preference measure was constructed to identify respondents that rated a single party (or coalition) higher than all the other parties (or coalitions). In the Austrian survey, similar questions were asked about seven plausible coalitions. A particular challenge is the measurement of accurate expectations. Various approaches have been used in the literature (see Blais et al., 2008). The most challenging approaches try to obtain precise numerical estimates, either of party vote shares or seats in parliament. These approaches might delight a political scientist, but it is rather unlikely that voters have (and should have) precise knowledge of these numbers. Some respondents will give impossible answers ( outliers ), forcing major adjustments or the exclusion of respondents from the analysis (e.g. Levine, 2007). In the context of a phone survey, even politically sophisticated respondents will

15 13 often fail to give precise numerical estimates for five or more parties that will add up to 100 percent (or know the precise number of seats in parliament). There are other and more reasonable ways of measuring expectations. For small parties in political systems with minimum vote thresholds, a more meaningful question asks respondents about the likelihood that a party will be able to pass the minimum vote threshold. For larger parties, the question can be posed as performance relative to a meaningful reference point such as the previous election result (more votes, fewer votes, or unchanged). Finally, the question of the election winner can be open, letting respondents define the winner as party, candidate, or coalition. In fact, the meaning of winning and losing can be ambiguous in multiparty systems with coalition governments (Hardmeier and Roth, 2003). The Austrian survey was especially designed to collect data for all these measures, while the German survey offers only a more limited subset. In order to evaluate the quality or accuracy of the electoral expectations, a plausible objective benchmark is needed. Published polls that were available while the survey was in the field are the obvious choice. In both countries, the polls did not fluctuate much during these periods. More specifically, for each forecast or judgment that was supported by the polls, a respondent would receive a point. For example, if a respondent thought it likely or certain that a party would obtain sufficient votes to pass the minimum vote threshold and the polls showed this party above the threshold, voter expectation and external poll matched. If a respondent made a wrong or no judgment at all, no point was given. In the Austrian survey, respondents answered questions about the electoral chances of six parties and seven plausible coalitions (whether they would have a majority after the election). All points were added to a summary score, ranging from 0 to 13, that indicates the overall accuracy of the forecasts. On average, 70 percent of these judgments were accurate. In the German survey, respondents were

16 14 asked similar questions about the three small parties and one question about a coalition. Here, the accuracy index ranges from 0 to 4. On average, respondents made correct predictions in 66 percent of the cases. 4. Results 4.1 Overall Quality of Electoral Expectations In a first step, the overall quality or ability of respondents to make accurate electoral predictions is investigated. The quality of the expectations was measured by the overall additive indicators of correct predictions as described above. The independent variables represent general political motivations, rational considerations, social context, and demographic control variables. Political motivations include party identification, interest in politics in general and the campaign in particular, self-reported attention to polls (Austria only), and political knowledge. Each factor is expected to improve the overall quality of the forecasts. Substantively most interesting is party identification, given its central role in wishful thinking about specific parties. If party identification distorts all party predictions, the effect on the overall quality index would be negative. If party identification captures predominately political interest and involvement, it should increase the overall predictive accuracy. Rational considerations are more difficult to operationalize. The indicators used here try to capture circumstances that would justify the costs and effort of acquiring accurate poll information. For example, voters with a clear preference for a single party are very unlikely to change their vote intention based on the expected outcome of the election. However, voters who prefer two or more parties to a similar degree should be more likely to consider the electoral chances of these parties. To maximize expected utility, strategic voters should decide their vote based on which party has the best chance of influencing the formation of the next government. The need for accurate expectations was operationalized as the evaluative distance of the parties

17 15 rated highest and second highest on the evaluation scales. A higher distance between these parties makes the decision between these parties easier. Consequently, evaluative distance should affect predictive accuracy negatively. In addition, the declared intention to vote is considered to be an indicator for the need for poll information (while controlling for political interest). Reporting the lack of an alternative choice to the declared vote intention (Austria only), however, should lower the need for accurate expectations. Both surveys do not provide direct measures of the partisan nature of respondents immediate social context or personal networks. Thus, the operationalization has to rely on regional differences. Respondents who live in regions or states that differ considerably from the national average might mistakenly distort their perceptions. To capture regional differences in Austria, the differences in vote shares at national and state levels for the five largest parties are added to form a single indicator of regional differences. For Germany, a single dichotomous indicator differentiates respondents in the Eastern part of Germany from respondents in the Western part of Germany, capturing the continuing and significant differences in voting behavior between both regions. In addition, a dichotomous indicator for respondents in the two capitals Vienna and Berlin is intended to capture the unique political conditions in each national capital. Both cities offer voters first-row seats to national politics. While Vienna is the only city state in Austria, Berlin was split between East and West during the cold war, making a clear regional assignment difficult. In short, these voters are expected to have a better grasp of political map. Political knowledge is operationalized with factual knowledge questions. In the Austrian survey, respondents were asked four questions (unemployment rate; name of at least one candidate in the regional electoral district; majority party in the Bundesrat, the second chamber of parliament; minimum vote threshold for the Nationalrat). The correct answers were added to a knowledge score that can range from 0 to 4. In the German survey, only a single factual

18 16 knowledge question was asked (the party with a majority in the second chamber of parliament, or Bundesrat). A correct answer was scored as 1. The two additive indicators of accurate forecasts were regressed on the predictor variables discussed above. The results provide support for all three types of explanatory factors (Table 1). Nearly all political motivations have significant positive effects on accurate perceptions, including party identification. Thus, the evidence supports the notion that, in general, partisan preferences are positively related to the ability to make accurate forecasts due to higher political motivation and involvement (while not excluding the possibility that partisan preferences lead to wishful thinking when it comes to specific liked or disliked parties). The positive effects of political knowledge and political interest were expected, but the marginal or complete lack of impact for interest in the political campaign is somewhat surprising. [Table 1 about here] The two social context variables tend to support the hypotheses. Those residing in Austrian states that differ from the national average, or in the Eastern part of Germany, do somewhat worse while respondents in the national capitals do somewhat better than the average. However, only respondents in Vienna outperform the rest of the country at marginally significant levels. The closeness to national politics seems to give these respondents a slight edge. Overall, the marginally significant negative effects for regional differences suggest that the prevalent national polls in the campaign coverage have not entirely supplanted more regional or local influences. As expected, political knowledge has strong positive effects on the quality of expectations. In fact, knowledge has the strongest effect of all the variables in the model, including political interest and education (beta coefficients not reported in table). This finding

19 17 confirms the central role of factual political knowledge for getting political information (Price and Zaller, 1993; Zaller, 1992). Among the demographic variables, the significant positive effect for male respondents in both countries is surprising. It should be noted that a similar effect was found for Dutch voters by Levine (2007). It is unclear why this apparently robust effect occurs in different settings and after controlling for various other variables. The main conclusion of this first part of the analyses is that political motivations, in particular partisan identification, improve rather than distort the overall quality of electoral expectations. 4.2 Wishful Thinking, Political Knowledge, and Expectations about Individual Parties and Coalitions The fact that partisan political motivations improve the overall quality of electoral forecasts does not preclude that expectations for individual parties and coalitions are affected or distorted by political preferences. To test whether Austrian and German respondents engaged in wishful thinking, the electoral expectations for parties and coalitions were regressed on the partisan preferences and evaluations, controlled for and moderated by political knowledge and education. Those who prefer and rate a party higher than all other parties should be more prone to wishful thinking and overestimate the electoral chances of these parties. The opposite should happen for disliked parties. Previous research suggests that wishful thinking is a highly robust effect (e.g. Blais and Turgeon, 2004; Granberg and Brent, 1983; Granberg and Holmberg, 1988; Babad and Yacobus, 1993; Mutz, 1998), and only education (Lewis-Beck and Skalaban, 1989) and political knowledge (Dolan and Holbrook, 2001) have been found to work against biased judgments and to reduce any wishful thinking effect.

20 18 In previous research, the wishful thinking effect was tested with predictions for individual parties. This test is both replicated and extended in the present analysis. First, wishful thinking effects were not estimated in separate models for each party, five in Austria and three in Germany. 1 For each country, the judgments were combined ( stacked ) and estimated in a single model. The expectations were regressed on the respective party preference, party evaluation, political knowledge, education, and the interactions between knowledge and education with the party evaluations. Second, the party model is replicated with expectations for coalitions. For Austria, the model includes the expectations for five different and plausible two-party coalitions. For Germany, the model includes three coalition expectations. 2 The independent variables are largely similar to the party model, except that party evaluations for both parties in a given coalition were included each model. The knowledge and education variables are interacted with the coalition preference indicator. Because each respondent contributes several (three to five) judgments to each data set, these judgments are not statistically independent from each other. Consequently, all judgments based on a single respondent were treated as a cluster and Table 2 reports robust standard errors, corrected for clustering (note that due to missing values, not all respondents contribute the exact same number of judgments). [Table 2 about here] 1 For Austria, the parties include the two large parties ÖVP and SPÖ as well as the three small parties FPÖ, BZÖ, and Greens. For Germany, expectations were only asked for three small parties, FDP, Greens, and Left Party. 2 For Austria, the coalitions include the grand coalition of ÖVP and SPÖ as well as the combinations ÖVP-FPÖ, ÖVP-BZÖ, ÖVP-Greens, and SPÖ-Greens. For Germany, the possible coalitions include CDU-FDP, SPD-Greens, and the grand coalition of CDU and SPD.

21 19 The results show strong evidence for wishful thinking. In both countries and for both party and coalition models, the preference for party or coalition has significant positive effects, indicating an overestimation of the electoral chances for preferred parties and coalitions. The party evaluations also contribute to wishful thinking, with the exception of the German coalition model. Here, the first coalition party evaluation has no effect at all and the second coalition party evaluation even has a negative sign. Knowledge and education have significant positive main effects. Given the electoral context in Austria and Germany, these effects can be explained by the fact that higher estimates of expected chances were generally accurate due to the fairly high chances of the small parties (that is, polls suggested that most small parties would be able to pass the minimum vote threshold). More important, however, are the interactions of knowledge and education with party evaluations and coalition preferences. The significant negative effects suggest that knowledge and education reduce the tendency of wishful thinking, confirming the findings of previous studies. Educated and knowledgeable respondents seem to have more and better information that allows them to limit the distorting effect of partisan preferences. Less knowledgeable respondents, on the other hand, might rely more on their partisan preferences and/or their (biased) social context, resulting in more distorted expectations. Overall, voters expectations about electoral chances in Austria and Germany were fairly accurate, and the evidence supports several explanations offered in the literature. The key contribution of our analysis is the careful distinction between overall accuracy of expectations and wishful thinking for individual parties and coalitions. Partisan preferences increase both, but the effect can arguably be considered positive in the first case and negative in the latter. 4.3 Expectations and Voting Behavior The literature review identified two particular mechanisms how expectations might influence voting behavior if the appropriate circumstances are given. In the rational and strategic

22 20 version, electoral expectations are used by rational voters to maximize expected utility by voting for the party with the highest chance of producing a desirable electoral outcome. This sophisticated mechanism was tested for votes for two small parties, the Left Party in Germany and the Alliance for the Future of Austria (BZÖ). The second and more heuristic mechanism, the bandwagon effect, will be tested with the vote intention for the ÖVP in Austria Strategic Voting: Avoiding Wasted Votes and Voting for Coalitions The general election in Germany provides a particularly interesting case for strategic voting, both related to the Left Party. The first and straightforward rational calculation is based on the wasted vote argument of strategic voting. A voter should only cast a ballot for a party if he or she is certain that the party has a realistic chance of passing the minimum vote threshold. If the party is certain to fail this threshold, a vote for this party would be wasted. The second and fairly sophisticated calculation is based on expectations about possible coalitions after the election (Linhart 2007). During the campaign, it was more or less certain that the Left Party would not become part of any coalition government (all other parties had explicitly rejected such a coalition). However, in case it would gain a substantial vote share, it was likely to preclude the formation of either a center-left coalition (the incumbent coalition of Social Democrats and Greens) or a center-right coalition (Christian Democrats and Free Democrats), forcing the formation of a grand coalition of SPD and CDU. In other words, if a voter preferred and expected a grand coalition while not expecting a left-of-center coalition of SPD, Greens, and Left Party (while controlling for party preferences for the Left Party), a vote for the Left Party can be considered to be rational. For most supporters of a grand coalition, a vote for the Left Party would probably be a vote for the least preferred party, but still a vote that would make the desired outcome of a grand coalition more likely.

23 21 To test these two decision strategies, the vote intention for the Left Party was regressed on indicators for expectations and preferences as discussed above as well as a dichotomous control variable for respondents in the Eastern part of Germany (and excluding those without a vote intention and those intending to abstain). The results support both strategic explanations (Table 3). After controlling for several preference measures for the Left Party (dichotomous party preference, party evaluation, strength of party identification, and the evaluation of party leader Oskar Lafontaine), the expectation that the Left Party will pass the minimum vote threshold still has a strong significant effect on the vote intention for the Left Party. In other words, those who do not expect the party to pass the threshold are more likely to avoid a wasted vote for a party without chances. The strategy of avoiding a wasted vote is supported primarily for respondents with high levels of political knowledge, but even the decisions of voters with a low level of knowledge are marginally affected by these expectations. [Table 3 about here] The second strategy of an indirect coalition vote was also supported. Respondents who preferred a grand coalition were more likely to vote for the Left Party. In contrast to the wasted vote decision, this rather sophisticated and complicated coalition vote is only supported for respondents with a high level of political knowledge. The first strategy of avoiding a wasted vote can be tested in Austria as well. Polls suggested that the Alliance for the Future of Austria (BZÖ), the junior coalition partner of the ÖVP in the incumbent coalition government, was in danger of failing the minimum vote threshold for seats in parliament. Under these circumstances, the expectations about the electoral chances of the party should matter above and beyond the relevant preference measures. Because

24 22 the BZÖ was highly unlikely to become part of the next government, coalition considerations do not seem plausible and are not included in the model. The vote intention for the BZÖ was regressed on the expectation of its electoral chances as well as a series of preference measures, including the evaluation of the well-known former (polarizing) party leader and governor of the state Carinthia at that time. Consequently, an additional indicator of the social context indicates whether a respondent resides in Carinthia. The results replicate the findings for Germany. The expectation of passing the minimum vote threshold again increases the likelihood of voting for the small party up to 28 percentage points for respondents with a BZÖ preference that strongly identify with their preferred party (Table 4). The wasted vote argument is supported again. 3 [Table 4 about here] The Bandwagon Effect The bandwagon effect was identified as the second mechanism of how expectations can affect voting behavior. Unlike strategic voting, the bandwagon effect does not require sophisticated decision making and rather relies on a simple heuristic cue, the perception of a likely winner. In multiparty systems with proportional representation and coalition governments, the concept of a winner is rather ambiguous because both large and small parties might rightfully make such a claim (Hardmeier and Roth 2003). The ultimate question, however, is which party, candidate, or coalition is perceived by voters as the winner. The Austrian survey can answer this question because it asked respondents an open question at the beginning of the 3 Separate tests for respondents with high and low levels of knowledge are not possible due to the lack of variance. Because relatively few respondents in the Austrian sample expressed a vote intention for the BZÖ, several variables do not vary within different levels of knowledge, precluding the estimation of these models.

25 23 interview: Who will win the upcoming election? Among the 1333 responses, 76 percent of respondents mentioned one of the two large parties (ÖVP or SPÖ), and 15 percent one of the two chancellor candidates of these two parties. Less than six percent of the respondents mentioned a coalition, and less than three percent a small party. In other words, more than 90 percent of the respondents perceived the 2006 Austrian election as a contest between the two large parties and/or their respective chancellor candidates. However, it is misleading to call this choice a contest because 76 percent of the respondents expected the ÖVP or its candidate to win. Only 17 percent of respondents expected the SPÖ to win. This pattern of responses allows a fairly straightforward test of a bandwagon effect. The (dichotomous) vote intention for the ÖVP is regressed on the expectation that the ÖVP or its candidate will win the election while controlling for various measures of partisan preferences for the ÖVP and government performance. A significant positive effect of the expectation variable would suggest a bandwagon effect. The results suggest that the expectation of the ÖVP winning the election has indeed a significant additional effect on the vote intention for the ÖVP, above and beyond various preference measures (Table 5). Unlike the coalition voting effect reported above, this bandwagon effect is not significant for voters with a high level of knowledge but only among respondents with a low level of knowledge. This differential effect suggests that the bandwagon effect is indeed rather an unsophisticated heuristic used primarily by voters without information for a more sophisticated decision. [Table 5 about here] 5. Discussion The paper tested and found support for a number of factors that affect voters ability to form accurate expectations about the electoral chances of parties and coalitions in two different countries. Political motivations and knowledge, rational considerations, and social context all

26 24 seem to contribute significantly to better and more accurate expectations. Most striking is the positive effect of partisan identifications on the overall quality of expectations, apparently contradicting previous research on wishful thinking. Because party identification is associated with higher political interest and involvement, the positive effect is not only very reasonable but also does not preclude wishful thinking effects for individual preferred or disliked parties. This is exactly what the results suggest. The analysis of individual expectations shows strong support for wishful thinking, not only for parties but for coalitions as well. Increasing levels of political knowledge and education seem to work against these biased perceptions and facilitate more accurate judgments. In the final part, two mechanisms how voters expectations might affect voting behavior were tested. First, two sophisticated strategic mechanisms found support. Both Austrian and German voters appear to avoid casting wasted votes for small parties that are not expected to pass the minimum vote threshold. Once electoral success is seen as certain, voters become more likely to cast a ballot in support of these parties. The wasted vote strategy was supported for voters with high but also low levels of political knowledge. The second and rather challenging mechanism of casting a strategic vote for a less (or least) preferred party in order to facilitate the formation of a desired coalition was supported for German voters as well. It is important to keep in mind that the test relies on circumstantial evidence, not direct measures of strategic considerations. Thus, the evidence is plausible but in need of further corroboration by more direct measures. The second mechanism, the bandwagon effect, is rather straightforward and supported for Austrian voters as well. Perceiving a party as a clear winner appears to increases the likelihood of casting a vote for this party, even in multiparty systems with proportional representation.